Local prediction with Saved Model

local_prediction_saved_model.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This is an example to make local predictions using the model in Saved Model format.\n",
    "\n",
    "In this example, you use the model from the babyweight tutorial:\n",
    "\n",
    "https://github.com/GoogleCloudPlatform/training-data-analyst/blob/master/blogs/babyweight/babyweight.ipynb"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.0.0\n"
     ]
    }
   ],
   "source": [
    "import tensorflow as tf\n",
    "print tf.__version__"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Suppose that you have trained the model locally on the datalab, and the model is exported under:\n",
    "\n",
    "`babyweight_trained/export/Servo/1501143087336`"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The following code imports the model and extract signatures from it."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "export_dir = 'babyweight_trained/export/Servo/1501143087336'\n",
    "\n",
    "sess = tf.Session()\n",
    "meta_graph = tf.saved_model.loader.load(sess, [tf.saved_model.tag_constants.SERVING], export_dir)\n",
    "model_signature = meta_graph.signature_def['serving_default']\n",
    "input_signature = model_signature.inputs\n",
    "output_signature = model_signature.outputs"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Check input and output key names."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[u'cigarette_use', u'gestation_weeks', u'is_male', u'mother_race', u'plurality', u'alcohol_use', u'mother_married', u'mother_age']\n",
      "[u'outputs']\n"
     ]
    }
   ],
   "source": [
    "print input_signature.keys()\n",
    "print output_signature.keys()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Each key has a correspoding Tensor in the graph and you can extract them using the following code:\n",
    "```\n",
    "sess.graph.get_tensor_by_name(input_signature['cigarette_use'].name)\n",
    "sess.graph.get_tensor_by_name(input_signature['gestation_weeks'].name)\n",
    "...\n",
    "sess.graph.get_tensor_by_name(output_signature['outputs'].name)\n",
    "```"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Here's an example to build a feed_dict dictionary and an output tensor which you can pass to the session."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "features = {\n",
    "      'is_male': 'True',\n",
    "      'mother_age': 26.0,\n",
    "      'mother_race': 'Asian Indian',\n",
    "      'plurality': 1.0,\n",
    "      'gestation_weeks': 39,\n",
    "      'mother_married': 'True',\n",
    "      'cigarette_use': 'False',\n",
    "      'alcohol_use': 'False'\n",
    "    }\n",
    "\n",
    "feed_dict = {\n",
    "  sess.graph.get_tensor_by_name(input_signature[key].name): [val]\n",
    "  for key, val in features.iteritems()\n",
    "}\n",
    "\n",
    "output = sess.graph.get_tensor_by_name(output_signature['outputs'].name)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now you can run the inference using them."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([ 6.39090729], dtype=float32)"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sess.run(output, feed_dict=feed_dict)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}